不同光谱植被指数反演冬小麦叶氮含量的敏感性研究

【目的】氮素是作物生长发育过程中最重要的营养元素之一,研究叶氮含量反演的有效光谱指标设置,为应用高光谱植被指数反演作物叶氮含量,以及作物的实时监测与精确诊断提供重要依据。【方法】以冬小麦为例,选取涵盖冬小麦全生育期不同覆盖程度225组冠层光谱与叶氮含量数据,通过遥感方法建立模型,模拟了不同光谱指标,即中心波长、信噪比和波段宽度对定量模型的影响,通过模型精度评价指标决定系数(coefficient of determination,R~2)、根均方差(root mean square error,RMSE)、平均绝对误差(mean absolute error,MAE)、平均相对误差(mean relative error,MRE)和显著性检验水平(P0.01)确定最优模型及最佳指标,分析光谱指标对叶氮含量定量模型反演的敏感性和有效性。【结果】反演冬小麦叶氮含量的最佳植被指数为MTCI_B,与实测叶氮含量的相关性最好(R~2=0.7674,RMSE=0.5511%,MAE=0.4625%,MRE=11.11个百分点,且P0.01),对应的最佳指标为中心波长420 nm、508 nm和405 nm,波段宽度1 nm,信噪比大于70 DB;高覆盖状况反演的最优指数为RVIinf_r(R~2=0.6739,RMSE=0.2964%,MAE=0.2851%,MRE=6.44个百分点,且P0.01),最优中心波长为826 nm和760 nm;低覆盖状况反演的最优指数为MTCI(R~2=0.8252,RMSE=0.4032%,MAE=0.4408%,MRE=12.22个百分点,且P0.01),最优中心波长为750 nm、693 nm和680 nm;应用最适于高低覆盖的植被指数RVIinf_r和MTCI构建的联合反演模型(R~2=0.9286,RMSE=0.3416%,MAE=0.2988%,MRE=7.16个百分点,且P0.01),明显优于最佳单一指数MTCI_B;模拟Hyperion和HJ1A-HSI传感器数据,联合反演模型精度(R~2为0.92—0.93,RMSE在0.37%—0.39%,MAE为0.285%左右,MRE约为7.00个百分点)明显优于单一植被指数反演精度(R~2为0.79—0.81,RMSE为0.63%—0.66%,MAE为0.455%左右,MRE约为10.90个百分点)。【结论】利用高光谱植被指数可有效实现作物叶氮含量反演,作物叶氮含量定量反演对不同光谱指标—中心波长、信噪比和波段宽度,具有较强敏感性。应用多指数联合反演模型,可显著提高反演精度,并且联合反演模型在不同高光谱传感器下有一定普适性。

Sensitivity of Different Spectral Vegetation Index for Estimating Winter Wheat Leaf Nitrogen

【Objective】Nitrogen is one of the most important nutrients in crop growth and development. The objective of this paper is to study the setting of effective index of leaf nitrogen content inversion in order to provide an important basis for the application of hyperspectral vegetation index of leaf nitrogen content estimation, and for real-time monitoring and accurate diagnosis of crops.【Method】A total of 225 groups of canopy reflectance and leaf nitrogen content data which covering the whole winter wheat growth period and under different levels of coverage, were collected to simulate different spectral index like different central wavelengths, SNR and band width indicators, and to analyze the influence of different observation pattern on quantitative models. And then, the indicators of accuracy evaluation, coefficient of determination, root mean square error, mean absolute error, mean relative error and P＜0.01 were used to select the optimal model and the best indicators, and the sensitivity and effectiveness of leaf nitrogen content quantitative models inversion were analyzed with different spectral indicators.【Result】MTCI_B was the best vegetation index for leaf nitrogen content inversion with the center wavelengths of 420 nm, 508 nm and 405 nm, band width of 1nm, SNR greater than 70 DB; the correlation with measured nitrogen content was preferably (R²=0.7674, RMSE=0.5511% , MAE=0.4625%, MRE=11.11 percentage points and P＜0.01). RVI_{inf_r} was the best index for inversion of high coverage with the optimal center wavelengths 826 nm and 760 nm (R²=0.6739, RMSE=0.2964%, MAE=0.2851%, MRE=6.44 percentage points and P＜0.01). MTCI was the best index for inversion of low coverage nitrogen (R²=0.8252, RMSE=0.4032%, MAE=0.4408%, MRE=12.22 percentage points and P＜0.01), corresponding to the optimal center wavelengths 750 nm, 693 nm and 680 nm. Using hyperspectral vegetation indexes RVI_{inf_r} and MTCI to build a joint inversion model, the model accuracy evaluation result (R²=0.9286, RMSE=0.3416%, MAE=0.2988%, MRE=7.16 percentage points and P＜0.01) was significantly better than the best single index MTCI_B. When the optimal model was used to simulate Hyperion and HJ1A-HSI data, the accuracy of the joint model (R² reached 0.92-0.93, RMSE were between 0.37%-0.39%) was better than the single vegetation index (R² were 0.79-0.81, RMSE were between 0.63%-0.66%).【Conclusion】A good estimation of crop leaf nitrogen content could be realized by using hyperspectral vegetation index, quantitative inversion of crop leaf nitrogen content had a strong sensitivity with different spectral indexes, center wavelength, SNR, and band width. Application of multi-exponential joint inversion model significantly improved the accuracy of the inversion. And the joint inversion model had a certain degree of universality in different hyperspectral sensors.